1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/pagemap.h>
8 #include <linux/time.h>
9 #include <linux/init.h>
10 #include <linux/string.h>
11 #include <linux/backing-dev.h>
12 #include <linux/falloc.h>
13 #include <linux/writeback.h>
14 #include <linux/compat.h>
15 #include <linux/slab.h>
16 #include <linux/btrfs.h>
17 #include <linux/uio.h>
18 #include <linux/iversion.h>
19 #include <linux/fsverity.h>
20 #include <linux/iomap.h>
23 #include "transaction.h"
24 #include "btrfs_inode.h"
25 #include "print-tree.h"
30 #include "compression.h"
31 #include "delalloc-space.h"
35 #include "accessors.h"
36 #include "extent-tree.h"
37 #include "file-item.h"
42 /* simple helper to fault in pages and copy. This should go away
43 * and be replaced with calls into generic code.
45 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
46 struct page **prepared_pages,
50 size_t total_copied = 0;
52 int offset = offset_in_page(pos);
54 while (write_bytes > 0) {
55 size_t count = min_t(size_t,
56 PAGE_SIZE - offset, write_bytes);
57 struct page *page = prepared_pages[pg];
59 * Copy data from userspace to the current page
61 copied = copy_page_from_iter_atomic(page, offset, count, i);
63 /* Flush processor's dcache for this page */
64 flush_dcache_page(page);
67 * if we get a partial write, we can end up with
68 * partially up to date pages. These add
69 * a lot of complexity, so make sure they don't
70 * happen by forcing this copy to be retried.
72 * The rest of the btrfs_file_write code will fall
73 * back to page at a time copies after we return 0.
75 if (unlikely(copied < count)) {
76 if (!PageUptodate(page)) {
77 iov_iter_revert(i, copied);
84 write_bytes -= copied;
85 total_copied += copied;
87 if (offset == PAGE_SIZE) {
96 * unlocks pages after btrfs_file_write is done with them
98 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
99 struct page **pages, size_t num_pages,
103 u64 block_start = round_down(pos, fs_info->sectorsize);
104 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
106 ASSERT(block_len <= U32_MAX);
107 for (i = 0; i < num_pages; i++) {
108 /* page checked is some magic around finding pages that
109 * have been modified without going through btrfs_set_page_dirty
110 * clear it here. There should be no need to mark the pages
111 * accessed as prepare_pages should have marked them accessed
112 * in prepare_pages via find_or_create_page()
114 btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
116 unlock_page(pages[i]);
122 * After btrfs_copy_from_user(), update the following things for delalloc:
123 * - Mark newly dirtied pages as DELALLOC in the io tree.
124 * Used to advise which range is to be written back.
125 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
126 * - Update inode size for past EOF write
128 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
129 size_t num_pages, loff_t pos, size_t write_bytes,
130 struct extent_state **cached, bool noreserve)
132 struct btrfs_fs_info *fs_info = inode->root->fs_info;
137 u64 end_of_last_block;
138 u64 end_pos = pos + write_bytes;
139 loff_t isize = i_size_read(&inode->vfs_inode);
140 unsigned int extra_bits = 0;
142 if (write_bytes == 0)
146 extra_bits |= EXTENT_NORESERVE;
148 start_pos = round_down(pos, fs_info->sectorsize);
149 num_bytes = round_up(write_bytes + pos - start_pos,
150 fs_info->sectorsize);
151 ASSERT(num_bytes <= U32_MAX);
153 end_of_last_block = start_pos + num_bytes - 1;
156 * The pages may have already been dirty, clear out old accounting so
157 * we can set things up properly
159 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
160 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
163 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
168 for (i = 0; i < num_pages; i++) {
169 struct page *p = pages[i];
171 btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
172 btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
173 btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
177 * we've only changed i_size in ram, and we haven't updated
178 * the disk i_size. There is no need to log the inode
182 i_size_write(&inode->vfs_inode, end_pos);
187 * this is very complex, but the basic idea is to drop all extents
188 * in the range start - end. hint_block is filled in with a block number
189 * that would be a good hint to the block allocator for this file.
191 * If an extent intersects the range but is not entirely inside the range
192 * it is either truncated or split. Anything entirely inside the range
193 * is deleted from the tree.
195 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
196 * to deal with that. We set the field 'bytes_found' of the arguments structure
197 * with the number of allocated bytes found in the target range, so that the
198 * caller can update the inode's number of bytes in an atomic way when
199 * replacing extents in a range to avoid races with stat(2).
201 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
202 struct btrfs_root *root, struct btrfs_inode *inode,
203 struct btrfs_drop_extents_args *args)
205 struct btrfs_fs_info *fs_info = root->fs_info;
206 struct extent_buffer *leaf;
207 struct btrfs_file_extent_item *fi;
208 struct btrfs_ref ref = { 0 };
209 struct btrfs_key key;
210 struct btrfs_key new_key;
211 u64 ino = btrfs_ino(inode);
212 u64 search_start = args->start;
215 u64 extent_offset = 0;
217 u64 last_end = args->start;
223 int modify_tree = -1;
226 struct btrfs_path *path = args->path;
228 args->bytes_found = 0;
229 args->extent_inserted = false;
231 /* Must always have a path if ->replace_extent is true */
232 ASSERT(!(args->replace_extent && !args->path));
235 path = btrfs_alloc_path();
242 if (args->drop_cache)
243 btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
245 if (args->start >= inode->disk_i_size && !args->replace_extent)
248 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
251 ret = btrfs_lookup_file_extent(trans, root, path, ino,
252 search_start, modify_tree);
255 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
256 leaf = path->nodes[0];
257 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
258 if (key.objectid == ino &&
259 key.type == BTRFS_EXTENT_DATA_KEY)
264 leaf = path->nodes[0];
265 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
267 ret = btrfs_next_leaf(root, path);
274 leaf = path->nodes[0];
278 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
280 if (key.objectid > ino)
282 if (WARN_ON_ONCE(key.objectid < ino) ||
283 key.type < BTRFS_EXTENT_DATA_KEY) {
288 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
291 fi = btrfs_item_ptr(leaf, path->slots[0],
292 struct btrfs_file_extent_item);
293 extent_type = btrfs_file_extent_type(leaf, fi);
295 if (extent_type == BTRFS_FILE_EXTENT_REG ||
296 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
297 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
298 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
299 extent_offset = btrfs_file_extent_offset(leaf, fi);
300 extent_end = key.offset +
301 btrfs_file_extent_num_bytes(leaf, fi);
302 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
303 extent_end = key.offset +
304 btrfs_file_extent_ram_bytes(leaf, fi);
311 * Don't skip extent items representing 0 byte lengths. They
312 * used to be created (bug) if while punching holes we hit
313 * -ENOSPC condition. So if we find one here, just ensure we
314 * delete it, otherwise we would insert a new file extent item
315 * with the same key (offset) as that 0 bytes length file
316 * extent item in the call to setup_items_for_insert() later
319 if (extent_end == key.offset && extent_end >= search_start) {
320 last_end = extent_end;
321 goto delete_extent_item;
324 if (extent_end <= search_start) {
330 search_start = max(key.offset, args->start);
331 if (recow || !modify_tree) {
333 btrfs_release_path(path);
338 * | - range to drop - |
339 * | -------- extent -------- |
341 if (args->start > key.offset && args->end < extent_end) {
343 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
348 memcpy(&new_key, &key, sizeof(new_key));
349 new_key.offset = args->start;
350 ret = btrfs_duplicate_item(trans, root, path,
352 if (ret == -EAGAIN) {
353 btrfs_release_path(path);
359 leaf = path->nodes[0];
360 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
361 struct btrfs_file_extent_item);
362 btrfs_set_file_extent_num_bytes(leaf, fi,
363 args->start - key.offset);
365 fi = btrfs_item_ptr(leaf, path->slots[0],
366 struct btrfs_file_extent_item);
368 extent_offset += args->start - key.offset;
369 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
370 btrfs_set_file_extent_num_bytes(leaf, fi,
371 extent_end - args->start);
372 btrfs_mark_buffer_dirty(trans, leaf);
374 if (update_refs && disk_bytenr > 0) {
375 btrfs_init_generic_ref(&ref,
376 BTRFS_ADD_DELAYED_REF,
377 disk_bytenr, num_bytes, 0,
378 root->root_key.objectid);
379 btrfs_init_data_ref(&ref,
380 root->root_key.objectid,
382 args->start - extent_offset,
384 ret = btrfs_inc_extent_ref(trans, &ref);
386 btrfs_abort_transaction(trans, ret);
390 key.offset = args->start;
393 * From here on out we will have actually dropped something, so
394 * last_end can be updated.
396 last_end = extent_end;
399 * | ---- range to drop ----- |
400 * | -------- extent -------- |
402 if (args->start <= key.offset && args->end < extent_end) {
403 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
408 memcpy(&new_key, &key, sizeof(new_key));
409 new_key.offset = args->end;
410 btrfs_set_item_key_safe(trans, path, &new_key);
412 extent_offset += args->end - key.offset;
413 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
414 btrfs_set_file_extent_num_bytes(leaf, fi,
415 extent_end - args->end);
416 btrfs_mark_buffer_dirty(trans, leaf);
417 if (update_refs && disk_bytenr > 0)
418 args->bytes_found += args->end - key.offset;
422 search_start = extent_end;
424 * | ---- range to drop ----- |
425 * | -------- extent -------- |
427 if (args->start > key.offset && args->end >= extent_end) {
429 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
434 btrfs_set_file_extent_num_bytes(leaf, fi,
435 args->start - key.offset);
436 btrfs_mark_buffer_dirty(trans, leaf);
437 if (update_refs && disk_bytenr > 0)
438 args->bytes_found += extent_end - args->start;
439 if (args->end == extent_end)
447 * | ---- range to drop ----- |
448 * | ------ extent ------ |
450 if (args->start <= key.offset && args->end >= extent_end) {
453 del_slot = path->slots[0];
456 BUG_ON(del_slot + del_nr != path->slots[0]);
461 extent_type == BTRFS_FILE_EXTENT_INLINE) {
462 args->bytes_found += extent_end - key.offset;
463 extent_end = ALIGN(extent_end,
464 fs_info->sectorsize);
465 } else if (update_refs && disk_bytenr > 0) {
466 btrfs_init_generic_ref(&ref,
467 BTRFS_DROP_DELAYED_REF,
468 disk_bytenr, num_bytes, 0,
469 root->root_key.objectid);
470 btrfs_init_data_ref(&ref,
471 root->root_key.objectid,
473 key.offset - extent_offset, 0,
475 ret = btrfs_free_extent(trans, &ref);
477 btrfs_abort_transaction(trans, ret);
480 args->bytes_found += extent_end - key.offset;
483 if (args->end == extent_end)
486 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
491 ret = btrfs_del_items(trans, root, path, del_slot,
494 btrfs_abort_transaction(trans, ret);
501 btrfs_release_path(path);
508 if (!ret && del_nr > 0) {
510 * Set path->slots[0] to first slot, so that after the delete
511 * if items are move off from our leaf to its immediate left or
512 * right neighbor leafs, we end up with a correct and adjusted
513 * path->slots[0] for our insertion (if args->replace_extent).
515 path->slots[0] = del_slot;
516 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
518 btrfs_abort_transaction(trans, ret);
521 leaf = path->nodes[0];
523 * If btrfs_del_items() was called, it might have deleted a leaf, in
524 * which case it unlocked our path, so check path->locks[0] matches a
527 if (!ret && args->replace_extent &&
528 path->locks[0] == BTRFS_WRITE_LOCK &&
529 btrfs_leaf_free_space(leaf) >=
530 sizeof(struct btrfs_item) + args->extent_item_size) {
533 key.type = BTRFS_EXTENT_DATA_KEY;
534 key.offset = args->start;
535 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
536 struct btrfs_key slot_key;
538 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
539 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
542 btrfs_setup_item_for_insert(trans, root, path, &key,
543 args->extent_item_size);
544 args->extent_inserted = true;
548 btrfs_free_path(path);
549 else if (!args->extent_inserted)
550 btrfs_release_path(path);
552 args->drop_end = found ? min(args->end, last_end) : args->end;
557 static int extent_mergeable(struct extent_buffer *leaf, int slot,
558 u64 objectid, u64 bytenr, u64 orig_offset,
559 u64 *start, u64 *end)
561 struct btrfs_file_extent_item *fi;
562 struct btrfs_key key;
565 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
568 btrfs_item_key_to_cpu(leaf, &key, slot);
569 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
572 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
573 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
574 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
575 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
576 btrfs_file_extent_compression(leaf, fi) ||
577 btrfs_file_extent_encryption(leaf, fi) ||
578 btrfs_file_extent_other_encoding(leaf, fi))
581 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
582 if ((*start && *start != key.offset) || (*end && *end != extent_end))
591 * Mark extent in the range start - end as written.
593 * This changes extent type from 'pre-allocated' to 'regular'. If only
594 * part of extent is marked as written, the extent will be split into
597 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
598 struct btrfs_inode *inode, u64 start, u64 end)
600 struct btrfs_root *root = inode->root;
601 struct extent_buffer *leaf;
602 struct btrfs_path *path;
603 struct btrfs_file_extent_item *fi;
604 struct btrfs_ref ref = { 0 };
605 struct btrfs_key key;
606 struct btrfs_key new_key;
618 u64 ino = btrfs_ino(inode);
620 path = btrfs_alloc_path();
627 key.type = BTRFS_EXTENT_DATA_KEY;
630 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
633 if (ret > 0 && path->slots[0] > 0)
636 leaf = path->nodes[0];
637 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
638 if (key.objectid != ino ||
639 key.type != BTRFS_EXTENT_DATA_KEY) {
641 btrfs_abort_transaction(trans, ret);
644 fi = btrfs_item_ptr(leaf, path->slots[0],
645 struct btrfs_file_extent_item);
646 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
648 btrfs_abort_transaction(trans, ret);
651 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
652 if (key.offset > start || extent_end < end) {
654 btrfs_abort_transaction(trans, ret);
658 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
659 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
660 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
661 memcpy(&new_key, &key, sizeof(new_key));
663 if (start == key.offset && end < extent_end) {
666 if (extent_mergeable(leaf, path->slots[0] - 1,
667 ino, bytenr, orig_offset,
668 &other_start, &other_end)) {
669 new_key.offset = end;
670 btrfs_set_item_key_safe(trans, path, &new_key);
671 fi = btrfs_item_ptr(leaf, path->slots[0],
672 struct btrfs_file_extent_item);
673 btrfs_set_file_extent_generation(leaf, fi,
675 btrfs_set_file_extent_num_bytes(leaf, fi,
677 btrfs_set_file_extent_offset(leaf, fi,
679 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
680 struct btrfs_file_extent_item);
681 btrfs_set_file_extent_generation(leaf, fi,
683 btrfs_set_file_extent_num_bytes(leaf, fi,
685 btrfs_mark_buffer_dirty(trans, leaf);
690 if (start > key.offset && end == extent_end) {
693 if (extent_mergeable(leaf, path->slots[0] + 1,
694 ino, bytenr, orig_offset,
695 &other_start, &other_end)) {
696 fi = btrfs_item_ptr(leaf, path->slots[0],
697 struct btrfs_file_extent_item);
698 btrfs_set_file_extent_num_bytes(leaf, fi,
700 btrfs_set_file_extent_generation(leaf, fi,
703 new_key.offset = start;
704 btrfs_set_item_key_safe(trans, path, &new_key);
706 fi = btrfs_item_ptr(leaf, path->slots[0],
707 struct btrfs_file_extent_item);
708 btrfs_set_file_extent_generation(leaf, fi,
710 btrfs_set_file_extent_num_bytes(leaf, fi,
712 btrfs_set_file_extent_offset(leaf, fi,
713 start - orig_offset);
714 btrfs_mark_buffer_dirty(trans, leaf);
719 while (start > key.offset || end < extent_end) {
720 if (key.offset == start)
723 new_key.offset = split;
724 ret = btrfs_duplicate_item(trans, root, path, &new_key);
725 if (ret == -EAGAIN) {
726 btrfs_release_path(path);
730 btrfs_abort_transaction(trans, ret);
734 leaf = path->nodes[0];
735 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
736 struct btrfs_file_extent_item);
737 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
738 btrfs_set_file_extent_num_bytes(leaf, fi,
741 fi = btrfs_item_ptr(leaf, path->slots[0],
742 struct btrfs_file_extent_item);
744 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
745 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
746 btrfs_set_file_extent_num_bytes(leaf, fi,
748 btrfs_mark_buffer_dirty(trans, leaf);
750 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
751 num_bytes, 0, root->root_key.objectid);
752 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
753 orig_offset, 0, false);
754 ret = btrfs_inc_extent_ref(trans, &ref);
756 btrfs_abort_transaction(trans, ret);
760 if (split == start) {
763 if (start != key.offset) {
765 btrfs_abort_transaction(trans, ret);
776 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
777 num_bytes, 0, root->root_key.objectid);
778 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
780 if (extent_mergeable(leaf, path->slots[0] + 1,
781 ino, bytenr, orig_offset,
782 &other_start, &other_end)) {
784 btrfs_release_path(path);
787 extent_end = other_end;
788 del_slot = path->slots[0] + 1;
790 ret = btrfs_free_extent(trans, &ref);
792 btrfs_abort_transaction(trans, ret);
798 if (extent_mergeable(leaf, path->slots[0] - 1,
799 ino, bytenr, orig_offset,
800 &other_start, &other_end)) {
802 btrfs_release_path(path);
805 key.offset = other_start;
806 del_slot = path->slots[0];
808 ret = btrfs_free_extent(trans, &ref);
810 btrfs_abort_transaction(trans, ret);
815 fi = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_file_extent_item);
817 btrfs_set_file_extent_type(leaf, fi,
818 BTRFS_FILE_EXTENT_REG);
819 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
820 btrfs_mark_buffer_dirty(trans, leaf);
822 fi = btrfs_item_ptr(leaf, del_slot - 1,
823 struct btrfs_file_extent_item);
824 btrfs_set_file_extent_type(leaf, fi,
825 BTRFS_FILE_EXTENT_REG);
826 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
827 btrfs_set_file_extent_num_bytes(leaf, fi,
828 extent_end - key.offset);
829 btrfs_mark_buffer_dirty(trans, leaf);
831 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
833 btrfs_abort_transaction(trans, ret);
838 btrfs_free_path(path);
843 * on error we return an unlocked page and the error value
844 * on success we return a locked page and 0
846 static int prepare_uptodate_page(struct inode *inode,
847 struct page *page, u64 pos,
850 struct folio *folio = page_folio(page);
853 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
854 !PageUptodate(page)) {
855 ret = btrfs_read_folio(NULL, folio);
859 if (!PageUptodate(page)) {
865 * Since btrfs_read_folio() will unlock the folio before it
866 * returns, there is a window where btrfs_release_folio() can be
867 * called to release the page. Here we check both inode
868 * mapping and PagePrivate() to make sure the page was not
871 * The private flag check is essential for subpage as we need
872 * to store extra bitmap using page->private.
874 if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
882 static fgf_t get_prepare_fgp_flags(bool nowait)
884 fgf_t fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
887 fgp_flags |= FGP_NOWAIT;
892 static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
896 gfp = btrfs_alloc_write_mask(inode->i_mapping);
898 gfp &= ~__GFP_DIRECT_RECLAIM;
906 * this just gets pages into the page cache and locks them down.
908 static noinline int prepare_pages(struct inode *inode, struct page **pages,
909 size_t num_pages, loff_t pos,
910 size_t write_bytes, bool force_uptodate,
914 unsigned long index = pos >> PAGE_SHIFT;
915 gfp_t mask = get_prepare_gfp_flags(inode, nowait);
916 fgf_t fgp_flags = get_prepare_fgp_flags(nowait);
920 for (i = 0; i < num_pages; i++) {
922 pages[i] = pagecache_get_page(inode->i_mapping, index + i,
923 fgp_flags, mask | __GFP_WRITE);
933 err = set_page_extent_mapped(pages[i]);
940 err = prepare_uptodate_page(inode, pages[i], pos,
942 if (!err && i == num_pages - 1)
943 err = prepare_uptodate_page(inode, pages[i],
944 pos + write_bytes, false);
947 if (!nowait && err == -EAGAIN) {
954 wait_on_page_writeback(pages[i]);
960 unlock_page(pages[faili]);
961 put_page(pages[faili]);
969 * This function locks the extent and properly waits for data=ordered extents
970 * to finish before allowing the pages to be modified if need.
973 * 1 - the extent is locked
974 * 0 - the extent is not locked, and everything is OK
975 * -EAGAIN - need re-prepare the pages
976 * the other < 0 number - Something wrong happens
979 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
980 size_t num_pages, loff_t pos,
982 u64 *lockstart, u64 *lockend, bool nowait,
983 struct extent_state **cached_state)
985 struct btrfs_fs_info *fs_info = inode->root->fs_info;
991 start_pos = round_down(pos, fs_info->sectorsize);
992 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
994 if (start_pos < inode->vfs_inode.i_size) {
995 struct btrfs_ordered_extent *ordered;
998 if (!try_lock_extent(&inode->io_tree, start_pos, last_pos,
1000 for (i = 0; i < num_pages; i++) {
1001 unlock_page(pages[i]);
1009 lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1012 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1013 last_pos - start_pos + 1);
1015 ordered->file_offset + ordered->num_bytes > start_pos &&
1016 ordered->file_offset <= last_pos) {
1017 unlock_extent(&inode->io_tree, start_pos, last_pos,
1019 for (i = 0; i < num_pages; i++) {
1020 unlock_page(pages[i]);
1023 btrfs_start_ordered_extent(ordered);
1024 btrfs_put_ordered_extent(ordered);
1028 btrfs_put_ordered_extent(ordered);
1030 *lockstart = start_pos;
1031 *lockend = last_pos;
1036 * We should be called after prepare_pages() which should have locked
1037 * all pages in the range.
1039 for (i = 0; i < num_pages; i++)
1040 WARN_ON(!PageLocked(pages[i]));
1046 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1048 * @pos: File offset.
1049 * @write_bytes: The length to write, will be updated to the nocow writeable
1052 * This function will flush ordered extents in the range to ensure proper
1056 * > 0 If we can nocow, and updates @write_bytes.
1057 * 0 If we can't do a nocow write.
1058 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1059 * root is in progress.
1060 * < 0 If an error happened.
1062 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1064 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1065 size_t *write_bytes, bool nowait)
1067 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1068 struct btrfs_root *root = inode->root;
1069 struct extent_state *cached_state = NULL;
1070 u64 lockstart, lockend;
1074 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1077 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1080 lockstart = round_down(pos, fs_info->sectorsize);
1081 lockend = round_up(pos + *write_bytes,
1082 fs_info->sectorsize) - 1;
1083 num_bytes = lockend - lockstart + 1;
1086 if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend,
1088 btrfs_drew_write_unlock(&root->snapshot_lock);
1092 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend,
1095 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1096 NULL, NULL, NULL, nowait, false);
1098 btrfs_drew_write_unlock(&root->snapshot_lock);
1100 *write_bytes = min_t(size_t, *write_bytes ,
1101 num_bytes - pos + lockstart);
1102 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
1107 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1109 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1112 static void update_time_for_write(struct inode *inode)
1114 struct timespec64 now, ts;
1116 if (IS_NOCMTIME(inode))
1119 now = current_time(inode);
1120 ts = inode_get_mtime(inode);
1121 if (!timespec64_equal(&ts, &now))
1122 inode_set_mtime_to_ts(inode, now);
1124 ts = inode_get_ctime(inode);
1125 if (!timespec64_equal(&ts, &now))
1126 inode_set_ctime_to_ts(inode, now);
1128 if (IS_I_VERSION(inode))
1129 inode_inc_iversion(inode);
1132 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1135 struct file *file = iocb->ki_filp;
1136 struct inode *inode = file_inode(file);
1137 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1138 loff_t pos = iocb->ki_pos;
1144 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1145 * prealloc flags, as without those flags we always have to COW. We will
1146 * later check if we can really COW into the target range (using
1147 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1149 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1150 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1153 ret = file_remove_privs(file);
1158 * We reserve space for updating the inode when we reserve space for the
1159 * extent we are going to write, so we will enospc out there. We don't
1160 * need to start yet another transaction to update the inode as we will
1161 * update the inode when we finish writing whatever data we write.
1163 update_time_for_write(inode);
1165 start_pos = round_down(pos, fs_info->sectorsize);
1166 oldsize = i_size_read(inode);
1167 if (start_pos > oldsize) {
1168 /* Expand hole size to cover write data, preventing empty gap */
1169 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1171 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1179 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1182 struct file *file = iocb->ki_filp;
1184 struct inode *inode = file_inode(file);
1185 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1186 struct page **pages = NULL;
1187 struct extent_changeset *data_reserved = NULL;
1188 u64 release_bytes = 0;
1191 size_t num_written = 0;
1194 bool only_release_metadata = false;
1195 bool force_page_uptodate = false;
1196 loff_t old_isize = i_size_read(inode);
1197 unsigned int ilock_flags = 0;
1198 const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1199 unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1202 ilock_flags |= BTRFS_ILOCK_TRY;
1204 ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1208 ret = generic_write_checks(iocb, i);
1212 ret = btrfs_write_check(iocb, i, ret);
1217 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1218 PAGE_SIZE / (sizeof(struct page *)));
1219 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1220 nrptrs = max(nrptrs, 8);
1221 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1227 while (iov_iter_count(i) > 0) {
1228 struct extent_state *cached_state = NULL;
1229 size_t offset = offset_in_page(pos);
1230 size_t sector_offset;
1231 size_t write_bytes = min(iov_iter_count(i),
1232 nrptrs * (size_t)PAGE_SIZE -
1235 size_t reserve_bytes;
1238 size_t dirty_sectors;
1243 * Fault pages before locking them in prepare_pages
1244 * to avoid recursive lock
1246 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1251 only_release_metadata = false;
1252 sector_offset = pos & (fs_info->sectorsize - 1);
1254 extent_changeset_release(data_reserved);
1255 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1256 &data_reserved, pos,
1257 write_bytes, nowait);
1261 if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1267 * If we don't have to COW at the offset, reserve
1268 * metadata only. write_bytes may get smaller than
1271 can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1272 &write_bytes, nowait);
1279 only_release_metadata = true;
1282 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1283 WARN_ON(num_pages > nrptrs);
1284 reserve_bytes = round_up(write_bytes + sector_offset,
1285 fs_info->sectorsize);
1286 WARN_ON(reserve_bytes == 0);
1287 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1289 reserve_bytes, nowait);
1291 if (!only_release_metadata)
1292 btrfs_free_reserved_data_space(BTRFS_I(inode),
1296 btrfs_check_nocow_unlock(BTRFS_I(inode));
1298 if (nowait && ret == -ENOSPC)
1303 release_bytes = reserve_bytes;
1305 ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1307 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1312 * This is going to setup the pages array with the number of
1313 * pages we want, so we don't really need to worry about the
1314 * contents of pages from loop to loop
1316 ret = prepare_pages(inode, pages, num_pages,
1317 pos, write_bytes, force_page_uptodate, false);
1319 btrfs_delalloc_release_extents(BTRFS_I(inode),
1324 extents_locked = lock_and_cleanup_extent_if_need(
1325 BTRFS_I(inode), pages,
1326 num_pages, pos, write_bytes, &lockstart,
1327 &lockend, nowait, &cached_state);
1328 if (extents_locked < 0) {
1329 if (!nowait && extents_locked == -EAGAIN)
1332 btrfs_delalloc_release_extents(BTRFS_I(inode),
1334 ret = extents_locked;
1338 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1340 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1341 dirty_sectors = round_up(copied + sector_offset,
1342 fs_info->sectorsize);
1343 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1346 * if we have trouble faulting in the pages, fall
1347 * back to one page at a time
1349 if (copied < write_bytes)
1353 force_page_uptodate = true;
1357 force_page_uptodate = false;
1358 dirty_pages = DIV_ROUND_UP(copied + offset,
1362 if (num_sectors > dirty_sectors) {
1363 /* release everything except the sectors we dirtied */
1364 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1365 if (only_release_metadata) {
1366 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1367 release_bytes, true);
1371 __pos = round_down(pos,
1372 fs_info->sectorsize) +
1373 (dirty_pages << PAGE_SHIFT);
1374 btrfs_delalloc_release_space(BTRFS_I(inode),
1375 data_reserved, __pos,
1376 release_bytes, true);
1380 release_bytes = round_up(copied + sector_offset,
1381 fs_info->sectorsize);
1383 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1384 dirty_pages, pos, copied,
1385 &cached_state, only_release_metadata);
1388 * If we have not locked the extent range, because the range's
1389 * start offset is >= i_size, we might still have a non-NULL
1390 * cached extent state, acquired while marking the extent range
1391 * as delalloc through btrfs_dirty_pages(). Therefore free any
1392 * possible cached extent state to avoid a memory leak.
1395 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1396 lockend, &cached_state);
1398 free_extent_state(cached_state);
1400 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1402 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1407 if (only_release_metadata)
1408 btrfs_check_nocow_unlock(BTRFS_I(inode));
1410 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1415 num_written += copied;
1420 if (release_bytes) {
1421 if (only_release_metadata) {
1422 btrfs_check_nocow_unlock(BTRFS_I(inode));
1423 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1424 release_bytes, true);
1426 btrfs_delalloc_release_space(BTRFS_I(inode),
1428 round_down(pos, fs_info->sectorsize),
1429 release_bytes, true);
1433 extent_changeset_free(data_reserved);
1434 if (num_written > 0) {
1435 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1436 iocb->ki_pos += num_written;
1439 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1440 return num_written ? num_written : ret;
1443 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1444 const struct iov_iter *iter, loff_t offset)
1446 const u32 blocksize_mask = fs_info->sectorsize - 1;
1448 if (offset & blocksize_mask)
1451 if (iov_iter_alignment(iter) & blocksize_mask)
1457 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1459 struct file *file = iocb->ki_filp;
1460 struct inode *inode = file_inode(file);
1461 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1463 ssize_t written = 0;
1464 ssize_t written_buffered;
1465 size_t prev_left = 0;
1468 unsigned int ilock_flags = 0;
1469 struct iomap_dio *dio;
1471 if (iocb->ki_flags & IOCB_NOWAIT)
1472 ilock_flags |= BTRFS_ILOCK_TRY;
1475 * If the write DIO is within EOF, use a shared lock and also only if
1476 * security bits will likely not be dropped by file_remove_privs() called
1477 * from btrfs_write_check(). Either will need to be rechecked after the
1478 * lock was acquired.
1480 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode) && IS_NOSEC(inode))
1481 ilock_flags |= BTRFS_ILOCK_SHARED;
1484 err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1488 /* Shared lock cannot be used with security bits set. */
1489 if ((ilock_flags & BTRFS_ILOCK_SHARED) && !IS_NOSEC(inode)) {
1490 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1491 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1495 err = generic_write_checks(iocb, from);
1497 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1501 err = btrfs_write_check(iocb, from, err);
1503 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1509 * Re-check since file size may have changed just before taking the
1510 * lock or pos may have changed because of O_APPEND in generic_write_check()
1512 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1513 pos + iov_iter_count(from) > i_size_read(inode)) {
1514 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1515 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1519 if (check_direct_IO(fs_info, from, pos)) {
1520 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1525 * The iov_iter can be mapped to the same file range we are writing to.
1526 * If that's the case, then we will deadlock in the iomap code, because
1527 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1528 * an ordered extent, and after that it will fault in the pages that the
1529 * iov_iter refers to. During the fault in we end up in the readahead
1530 * pages code (starting at btrfs_readahead()), which will lock the range,
1531 * find that ordered extent and then wait for it to complete (at
1532 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1533 * obviously the ordered extent can never complete as we didn't submit
1534 * yet the respective bio(s). This always happens when the buffer is
1535 * memory mapped to the same file range, since the iomap DIO code always
1536 * invalidates pages in the target file range (after starting and waiting
1537 * for any writeback).
1539 * So here we disable page faults in the iov_iter and then retry if we
1540 * got -EFAULT, faulting in the pages before the retry.
1542 from->nofault = true;
1543 dio = btrfs_dio_write(iocb, from, written);
1544 from->nofault = false;
1547 * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1548 * iocb, and that needs to lock the inode. So unlock it before calling
1549 * iomap_dio_complete() to avoid a deadlock.
1551 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1553 if (IS_ERR_OR_NULL(dio))
1554 err = PTR_ERR_OR_ZERO(dio);
1556 err = iomap_dio_complete(dio);
1558 /* No increment (+=) because iomap returns a cumulative value. */
1562 if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1563 const size_t left = iov_iter_count(from);
1565 * We have more data left to write. Try to fault in as many as
1566 * possible of the remainder pages and retry. We do this without
1567 * releasing and locking again the inode, to prevent races with
1570 * Also, in case the iov refers to pages in the file range of the
1571 * file we want to write to (due to a mmap), we could enter an
1572 * infinite loop if we retry after faulting the pages in, since
1573 * iomap will invalidate any pages in the range early on, before
1574 * it tries to fault in the pages of the iov. So we keep track of
1575 * how much was left of iov in the previous EFAULT and fallback
1576 * to buffered IO in case we haven't made any progress.
1578 if (left == prev_left) {
1581 fault_in_iov_iter_readable(from, left);
1588 * If 'err' is -ENOTBLK or we have not written all data, then it means
1589 * we must fallback to buffered IO.
1591 if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1596 * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1597 * it must retry the operation in a context where blocking is acceptable,
1598 * because even if we end up not blocking during the buffered IO attempt
1599 * below, we will block when flushing and waiting for the IO.
1601 if (iocb->ki_flags & IOCB_NOWAIT) {
1607 written_buffered = btrfs_buffered_write(iocb, from);
1608 if (written_buffered < 0) {
1609 err = written_buffered;
1613 * Ensure all data is persisted. We want the next direct IO read to be
1614 * able to read what was just written.
1616 endbyte = pos + written_buffered - 1;
1617 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1620 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1623 written += written_buffered;
1624 iocb->ki_pos = pos + written_buffered;
1625 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1626 endbyte >> PAGE_SHIFT);
1628 return err < 0 ? err : written;
1631 static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1632 const struct btrfs_ioctl_encoded_io_args *encoded)
1634 struct file *file = iocb->ki_filp;
1635 struct inode *inode = file_inode(file);
1639 btrfs_inode_lock(BTRFS_I(inode), 0);
1640 count = encoded->len;
1641 ret = generic_write_checks_count(iocb, &count);
1642 if (ret == 0 && count != encoded->len) {
1644 * The write got truncated by generic_write_checks_count(). We
1645 * can't do a partial encoded write.
1649 if (ret || encoded->len == 0)
1652 ret = btrfs_write_check(iocb, from, encoded->len);
1656 ret = btrfs_do_encoded_write(iocb, from, encoded);
1658 btrfs_inode_unlock(BTRFS_I(inode), 0);
1662 ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1663 const struct btrfs_ioctl_encoded_io_args *encoded)
1665 struct file *file = iocb->ki_filp;
1666 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1667 ssize_t num_written, num_sync;
1670 * If the fs flips readonly due to some impossible error, although we
1671 * have opened a file as writable, we have to stop this write operation
1672 * to ensure consistency.
1674 if (BTRFS_FS_ERROR(inode->root->fs_info))
1677 if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1681 num_written = btrfs_encoded_write(iocb, from, encoded);
1682 num_sync = encoded->len;
1683 } else if (iocb->ki_flags & IOCB_DIRECT) {
1684 num_written = btrfs_direct_write(iocb, from);
1685 num_sync = num_written;
1687 num_written = btrfs_buffered_write(iocb, from);
1688 num_sync = num_written;
1691 btrfs_set_inode_last_sub_trans(inode);
1694 num_sync = generic_write_sync(iocb, num_sync);
1696 num_written = num_sync;
1702 static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1704 return btrfs_do_write_iter(iocb, from, NULL);
1707 int btrfs_release_file(struct inode *inode, struct file *filp)
1709 struct btrfs_file_private *private = filp->private_data;
1712 kfree(private->filldir_buf);
1713 free_extent_state(private->llseek_cached_state);
1715 filp->private_data = NULL;
1719 * Set by setattr when we are about to truncate a file from a non-zero
1720 * size to a zero size. This tries to flush down new bytes that may
1721 * have been written if the application were using truncate to replace
1724 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1725 &BTRFS_I(inode)->runtime_flags))
1726 filemap_flush(inode->i_mapping);
1730 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1733 struct blk_plug plug;
1736 * This is only called in fsync, which would do synchronous writes, so
1737 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1738 * multiple disks using raid profile, a large IO can be split to
1739 * several segments of stripe length (currently 64K).
1741 blk_start_plug(&plug);
1742 ret = btrfs_fdatawrite_range(inode, start, end);
1743 blk_finish_plug(&plug);
1748 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1750 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
1751 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1753 if (btrfs_inode_in_log(inode, btrfs_get_fs_generation(fs_info)) &&
1754 list_empty(&ctx->ordered_extents))
1758 * If we are doing a fast fsync we can not bail out if the inode's
1759 * last_trans is <= then the last committed transaction, because we only
1760 * update the last_trans of the inode during ordered extent completion,
1761 * and for a fast fsync we don't wait for that, we only wait for the
1762 * writeback to complete.
1764 if (inode->last_trans <= btrfs_get_last_trans_committed(fs_info) &&
1765 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1766 list_empty(&ctx->ordered_extents)))
1773 * fsync call for both files and directories. This logs the inode into
1774 * the tree log instead of forcing full commits whenever possible.
1776 * It needs to call filemap_fdatawait so that all ordered extent updates are
1777 * in the metadata btree are up to date for copying to the log.
1779 * It drops the inode mutex before doing the tree log commit. This is an
1780 * important optimization for directories because holding the mutex prevents
1781 * new operations on the dir while we write to disk.
1783 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1785 struct dentry *dentry = file_dentry(file);
1786 struct inode *inode = d_inode(dentry);
1787 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1788 struct btrfs_root *root = BTRFS_I(inode)->root;
1789 struct btrfs_trans_handle *trans;
1790 struct btrfs_log_ctx ctx;
1795 trace_btrfs_sync_file(file, datasync);
1797 btrfs_init_log_ctx(&ctx, inode);
1800 * Always set the range to a full range, otherwise we can get into
1801 * several problems, from missing file extent items to represent holes
1802 * when not using the NO_HOLES feature, to log tree corruption due to
1803 * races between hole detection during logging and completion of ordered
1804 * extents outside the range, to missing checksums due to ordered extents
1805 * for which we flushed only a subset of their pages.
1809 len = (u64)LLONG_MAX + 1;
1812 * We write the dirty pages in the range and wait until they complete
1813 * out of the ->i_mutex. If so, we can flush the dirty pages by
1814 * multi-task, and make the performance up. See
1815 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1817 ret = start_ordered_ops(inode, start, end);
1821 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1823 atomic_inc(&root->log_batch);
1826 * Before we acquired the inode's lock and the mmap lock, someone may
1827 * have dirtied more pages in the target range. We need to make sure
1828 * that writeback for any such pages does not start while we are logging
1829 * the inode, because if it does, any of the following might happen when
1830 * we are not doing a full inode sync:
1832 * 1) We log an extent after its writeback finishes but before its
1833 * checksums are added to the csum tree, leading to -EIO errors
1834 * when attempting to read the extent after a log replay.
1836 * 2) We can end up logging an extent before its writeback finishes.
1837 * Therefore after the log replay we will have a file extent item
1838 * pointing to an unwritten extent (and no data checksums as well).
1840 * So trigger writeback for any eventual new dirty pages and then we
1841 * wait for all ordered extents to complete below.
1843 ret = start_ordered_ops(inode, start, end);
1845 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1850 * Always check for the full sync flag while holding the inode's lock,
1851 * to avoid races with other tasks. The flag must be either set all the
1852 * time during logging or always off all the time while logging.
1853 * We check the flag here after starting delalloc above, because when
1854 * running delalloc the full sync flag may be set if we need to drop
1855 * extra extent map ranges due to temporary memory allocation failures.
1857 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1858 &BTRFS_I(inode)->runtime_flags);
1861 * We have to do this here to avoid the priority inversion of waiting on
1862 * IO of a lower priority task while holding a transaction open.
1864 * For a full fsync we wait for the ordered extents to complete while
1865 * for a fast fsync we wait just for writeback to complete, and then
1866 * attach the ordered extents to the transaction so that a transaction
1867 * commit waits for their completion, to avoid data loss if we fsync,
1868 * the current transaction commits before the ordered extents complete
1869 * and a power failure happens right after that.
1871 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1872 * logical address recorded in the ordered extent may change. We need
1873 * to wait for the IO to stabilize the logical address.
1875 if (full_sync || btrfs_is_zoned(fs_info)) {
1876 ret = btrfs_wait_ordered_range(inode, start, len);
1879 * Get our ordered extents as soon as possible to avoid doing
1880 * checksum lookups in the csum tree, and use instead the
1881 * checksums attached to the ordered extents.
1883 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
1884 &ctx.ordered_extents);
1885 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
1889 goto out_release_extents;
1891 atomic_inc(&root->log_batch);
1893 if (skip_inode_logging(&ctx)) {
1895 * We've had everything committed since the last time we were
1896 * modified so clear this flag in case it was set for whatever
1897 * reason, it's no longer relevant.
1899 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1900 &BTRFS_I(inode)->runtime_flags);
1902 * An ordered extent might have started before and completed
1903 * already with io errors, in which case the inode was not
1904 * updated and we end up here. So check the inode's mapping
1905 * for any errors that might have happened since we last
1906 * checked called fsync.
1908 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
1909 goto out_release_extents;
1913 * We use start here because we will need to wait on the IO to complete
1914 * in btrfs_sync_log, which could require joining a transaction (for
1915 * example checking cross references in the nocow path). If we use join
1916 * here we could get into a situation where we're waiting on IO to
1917 * happen that is blocked on a transaction trying to commit. With start
1918 * we inc the extwriter counter, so we wait for all extwriters to exit
1919 * before we start blocking joiners. This comment is to keep somebody
1920 * from thinking they are super smart and changing this to
1921 * btrfs_join_transaction *cough*Josef*cough*.
1923 trans = btrfs_start_transaction(root, 0);
1924 if (IS_ERR(trans)) {
1925 ret = PTR_ERR(trans);
1926 goto out_release_extents;
1928 trans->in_fsync = true;
1930 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1931 btrfs_release_log_ctx_extents(&ctx);
1933 /* Fallthrough and commit/free transaction. */
1934 ret = BTRFS_LOG_FORCE_COMMIT;
1937 /* we've logged all the items and now have a consistent
1938 * version of the file in the log. It is possible that
1939 * someone will come in and modify the file, but that's
1940 * fine because the log is consistent on disk, and we
1941 * have references to all of the file's extents
1943 * It is possible that someone will come in and log the
1944 * file again, but that will end up using the synchronization
1945 * inside btrfs_sync_log to keep things safe.
1947 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1949 if (ret == BTRFS_NO_LOG_SYNC) {
1950 ret = btrfs_end_transaction(trans);
1954 /* We successfully logged the inode, attempt to sync the log. */
1956 ret = btrfs_sync_log(trans, root, &ctx);
1958 ret = btrfs_end_transaction(trans);
1964 * At this point we need to commit the transaction because we had
1965 * btrfs_need_log_full_commit() or some other error.
1967 * If we didn't do a full sync we have to stop the trans handle, wait on
1968 * the ordered extents, start it again and commit the transaction. If
1969 * we attempt to wait on the ordered extents here we could deadlock with
1970 * something like fallocate() that is holding the extent lock trying to
1971 * start a transaction while some other thread is trying to commit the
1972 * transaction while we (fsync) are currently holding the transaction
1976 ret = btrfs_end_transaction(trans);
1979 ret = btrfs_wait_ordered_range(inode, start, len);
1984 * This is safe to use here because we're only interested in
1985 * making sure the transaction that had the ordered extents is
1986 * committed. We aren't waiting on anything past this point,
1987 * we're purely getting the transaction and committing it.
1989 trans = btrfs_attach_transaction_barrier(root);
1990 if (IS_ERR(trans)) {
1991 ret = PTR_ERR(trans);
1994 * We committed the transaction and there's no currently
1995 * running transaction, this means everything we care
1996 * about made it to disk and we are done.
2004 ret = btrfs_commit_transaction(trans);
2006 ASSERT(list_empty(&ctx.list));
2007 ASSERT(list_empty(&ctx.conflict_inodes));
2008 err = file_check_and_advance_wb_err(file);
2011 return ret > 0 ? -EIO : ret;
2013 out_release_extents:
2014 btrfs_release_log_ctx_extents(&ctx);
2015 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2019 static const struct vm_operations_struct btrfs_file_vm_ops = {
2020 .fault = filemap_fault,
2021 .map_pages = filemap_map_pages,
2022 .page_mkwrite = btrfs_page_mkwrite,
2025 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2027 struct address_space *mapping = filp->f_mapping;
2029 if (!mapping->a_ops->read_folio)
2032 file_accessed(filp);
2033 vma->vm_ops = &btrfs_file_vm_ops;
2038 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2039 int slot, u64 start, u64 end)
2041 struct btrfs_file_extent_item *fi;
2042 struct btrfs_key key;
2044 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2047 btrfs_item_key_to_cpu(leaf, &key, slot);
2048 if (key.objectid != btrfs_ino(inode) ||
2049 key.type != BTRFS_EXTENT_DATA_KEY)
2052 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2054 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2057 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2060 if (key.offset == end)
2062 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2067 static int fill_holes(struct btrfs_trans_handle *trans,
2068 struct btrfs_inode *inode,
2069 struct btrfs_path *path, u64 offset, u64 end)
2071 struct btrfs_fs_info *fs_info = trans->fs_info;
2072 struct btrfs_root *root = inode->root;
2073 struct extent_buffer *leaf;
2074 struct btrfs_file_extent_item *fi;
2075 struct extent_map *hole_em;
2076 struct btrfs_key key;
2079 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2082 key.objectid = btrfs_ino(inode);
2083 key.type = BTRFS_EXTENT_DATA_KEY;
2084 key.offset = offset;
2086 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2089 * We should have dropped this offset, so if we find it then
2090 * something has gone horribly wrong.
2097 leaf = path->nodes[0];
2098 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2102 fi = btrfs_item_ptr(leaf, path->slots[0],
2103 struct btrfs_file_extent_item);
2104 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2106 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2107 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2108 btrfs_set_file_extent_offset(leaf, fi, 0);
2109 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2110 btrfs_mark_buffer_dirty(trans, leaf);
2114 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2117 key.offset = offset;
2118 btrfs_set_item_key_safe(trans, path, &key);
2119 fi = btrfs_item_ptr(leaf, path->slots[0],
2120 struct btrfs_file_extent_item);
2121 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2123 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2124 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2125 btrfs_set_file_extent_offset(leaf, fi, 0);
2126 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2127 btrfs_mark_buffer_dirty(trans, leaf);
2130 btrfs_release_path(path);
2132 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2138 btrfs_release_path(path);
2140 hole_em = alloc_extent_map();
2142 btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2143 btrfs_set_inode_full_sync(inode);
2145 hole_em->start = offset;
2146 hole_em->len = end - offset;
2147 hole_em->ram_bytes = hole_em->len;
2148 hole_em->orig_start = offset;
2150 hole_em->block_start = EXTENT_MAP_HOLE;
2151 hole_em->block_len = 0;
2152 hole_em->orig_block_len = 0;
2153 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2154 hole_em->generation = trans->transid;
2156 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2157 free_extent_map(hole_em);
2159 btrfs_set_inode_full_sync(inode);
2166 * Find a hole extent on given inode and change start/len to the end of hole
2167 * extent.(hole/vacuum extent whose em->start <= start &&
2168 * em->start + em->len > start)
2169 * When a hole extent is found, return 1 and modify start/len.
2171 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2173 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2174 struct extent_map *em;
2177 em = btrfs_get_extent(inode, NULL, 0,
2178 round_down(*start, fs_info->sectorsize),
2179 round_up(*len, fs_info->sectorsize));
2183 /* Hole or vacuum extent(only exists in no-hole mode) */
2184 if (em->block_start == EXTENT_MAP_HOLE) {
2186 *len = em->start + em->len > *start + *len ?
2187 0 : *start + *len - em->start - em->len;
2188 *start = em->start + em->len;
2190 free_extent_map(em);
2194 static void btrfs_punch_hole_lock_range(struct inode *inode,
2195 const u64 lockstart,
2197 struct extent_state **cached_state)
2200 * For subpage case, if the range is not at page boundary, we could
2201 * have pages at the leading/tailing part of the range.
2202 * This could lead to dead loop since filemap_range_has_page()
2203 * will always return true.
2204 * So here we need to do extra page alignment for
2205 * filemap_range_has_page().
2207 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2208 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2211 truncate_pagecache_range(inode, lockstart, lockend);
2213 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2216 * We can't have ordered extents in the range, nor dirty/writeback
2217 * pages, because we have locked the inode's VFS lock in exclusive
2218 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2219 * we have flushed all delalloc in the range and we have waited
2220 * for any ordered extents in the range to complete.
2221 * We can race with anyone reading pages from this range, so after
2222 * locking the range check if we have pages in the range, and if
2223 * we do, unlock the range and retry.
2225 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2229 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2233 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2236 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2237 struct btrfs_inode *inode,
2238 struct btrfs_path *path,
2239 struct btrfs_replace_extent_info *extent_info,
2240 const u64 replace_len,
2241 const u64 bytes_to_drop)
2243 struct btrfs_fs_info *fs_info = trans->fs_info;
2244 struct btrfs_root *root = inode->root;
2245 struct btrfs_file_extent_item *extent;
2246 struct extent_buffer *leaf;
2247 struct btrfs_key key;
2249 struct btrfs_ref ref = { 0 };
2252 if (replace_len == 0)
2255 if (extent_info->disk_offset == 0 &&
2256 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2257 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2261 key.objectid = btrfs_ino(inode);
2262 key.type = BTRFS_EXTENT_DATA_KEY;
2263 key.offset = extent_info->file_offset;
2264 ret = btrfs_insert_empty_item(trans, root, path, &key,
2265 sizeof(struct btrfs_file_extent_item));
2268 leaf = path->nodes[0];
2269 slot = path->slots[0];
2270 write_extent_buffer(leaf, extent_info->extent_buf,
2271 btrfs_item_ptr_offset(leaf, slot),
2272 sizeof(struct btrfs_file_extent_item));
2273 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2274 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2275 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2276 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2277 if (extent_info->is_new_extent)
2278 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2279 btrfs_mark_buffer_dirty(trans, leaf);
2280 btrfs_release_path(path);
2282 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2287 /* If it's a hole, nothing more needs to be done. */
2288 if (extent_info->disk_offset == 0) {
2289 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2293 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2295 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2296 key.objectid = extent_info->disk_offset;
2297 key.type = BTRFS_EXTENT_ITEM_KEY;
2298 key.offset = extent_info->disk_len;
2299 ret = btrfs_alloc_reserved_file_extent(trans, root,
2301 extent_info->file_offset,
2302 extent_info->qgroup_reserved,
2307 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2308 extent_info->disk_offset,
2309 extent_info->disk_len, 0,
2310 root->root_key.objectid);
2311 ref_offset = extent_info->file_offset - extent_info->data_offset;
2312 btrfs_init_data_ref(&ref, root->root_key.objectid,
2313 btrfs_ino(inode), ref_offset, 0, false);
2314 ret = btrfs_inc_extent_ref(trans, &ref);
2317 extent_info->insertions++;
2323 * The respective range must have been previously locked, as well as the inode.
2324 * The end offset is inclusive (last byte of the range).
2325 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2326 * the file range with an extent.
2327 * When not punching a hole, we don't want to end up in a state where we dropped
2328 * extents without inserting a new one, so we must abort the transaction to avoid
2331 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2332 struct btrfs_path *path, const u64 start,
2334 struct btrfs_replace_extent_info *extent_info,
2335 struct btrfs_trans_handle **trans_out)
2337 struct btrfs_drop_extents_args drop_args = { 0 };
2338 struct btrfs_root *root = inode->root;
2339 struct btrfs_fs_info *fs_info = root->fs_info;
2340 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2341 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2342 struct btrfs_trans_handle *trans = NULL;
2343 struct btrfs_block_rsv *rsv;
2344 unsigned int rsv_count;
2346 u64 len = end - start;
2352 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2357 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2358 rsv->failfast = true;
2361 * 1 - update the inode
2362 * 1 - removing the extents in the range
2363 * 1 - adding the hole extent if no_holes isn't set or if we are
2364 * replacing the range with a new extent
2366 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2371 trans = btrfs_start_transaction(root, rsv_count);
2372 if (IS_ERR(trans)) {
2373 ret = PTR_ERR(trans);
2378 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2382 trans->block_rsv = rsv;
2385 drop_args.path = path;
2386 drop_args.end = end + 1;
2387 drop_args.drop_cache = true;
2388 while (cur_offset < end) {
2389 drop_args.start = cur_offset;
2390 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2391 /* If we are punching a hole decrement the inode's byte count */
2393 btrfs_update_inode_bytes(inode, 0,
2394 drop_args.bytes_found);
2395 if (ret != -ENOSPC) {
2397 * The only time we don't want to abort is if we are
2398 * attempting to clone a partial inline extent, in which
2399 * case we'll get EOPNOTSUPP. However if we aren't
2400 * clone we need to abort no matter what, because if we
2401 * got EOPNOTSUPP via prealloc then we messed up and
2405 (ret != -EOPNOTSUPP ||
2406 (extent_info && extent_info->is_new_extent)))
2407 btrfs_abort_transaction(trans, ret);
2411 trans->block_rsv = &fs_info->trans_block_rsv;
2413 if (!extent_info && cur_offset < drop_args.drop_end &&
2414 cur_offset < ino_size) {
2415 ret = fill_holes(trans, inode, path, cur_offset,
2416 drop_args.drop_end);
2419 * If we failed then we didn't insert our hole
2420 * entries for the area we dropped, so now the
2421 * fs is corrupted, so we must abort the
2424 btrfs_abort_transaction(trans, ret);
2427 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2429 * We are past the i_size here, but since we didn't
2430 * insert holes we need to clear the mapped area so we
2431 * know to not set disk_i_size in this area until a new
2432 * file extent is inserted here.
2434 ret = btrfs_inode_clear_file_extent_range(inode,
2436 drop_args.drop_end - cur_offset);
2439 * We couldn't clear our area, so we could
2440 * presumably adjust up and corrupt the fs, so
2443 btrfs_abort_transaction(trans, ret);
2449 drop_args.drop_end > extent_info->file_offset) {
2450 u64 replace_len = drop_args.drop_end -
2451 extent_info->file_offset;
2453 ret = btrfs_insert_replace_extent(trans, inode, path,
2454 extent_info, replace_len,
2455 drop_args.bytes_found);
2457 btrfs_abort_transaction(trans, ret);
2460 extent_info->data_len -= replace_len;
2461 extent_info->data_offset += replace_len;
2462 extent_info->file_offset += replace_len;
2466 * We are releasing our handle on the transaction, balance the
2467 * dirty pages of the btree inode and flush delayed items, and
2468 * then get a new transaction handle, which may now point to a
2469 * new transaction in case someone else may have committed the
2470 * transaction we used to replace/drop file extent items. So
2471 * bump the inode's iversion and update mtime and ctime except
2472 * if we are called from a dedupe context. This is because a
2473 * power failure/crash may happen after the transaction is
2474 * committed and before we finish replacing/dropping all the
2475 * file extent items we need.
2477 inode_inc_iversion(&inode->vfs_inode);
2479 if (!extent_info || extent_info->update_times)
2480 inode_set_mtime_to_ts(&inode->vfs_inode,
2481 inode_set_ctime_current(&inode->vfs_inode));
2483 ret = btrfs_update_inode(trans, inode);
2487 btrfs_end_transaction(trans);
2488 btrfs_btree_balance_dirty(fs_info);
2490 trans = btrfs_start_transaction(root, rsv_count);
2491 if (IS_ERR(trans)) {
2492 ret = PTR_ERR(trans);
2497 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2498 rsv, min_size, false);
2501 trans->block_rsv = rsv;
2503 cur_offset = drop_args.drop_end;
2504 len = end - cur_offset;
2505 if (!extent_info && len) {
2506 ret = find_first_non_hole(inode, &cur_offset, &len);
2507 if (unlikely(ret < 0))
2517 * If we were cloning, force the next fsync to be a full one since we
2518 * we replaced (or just dropped in the case of cloning holes when
2519 * NO_HOLES is enabled) file extent items and did not setup new extent
2520 * maps for the replacement extents (or holes).
2522 if (extent_info && !extent_info->is_new_extent)
2523 btrfs_set_inode_full_sync(inode);
2528 trans->block_rsv = &fs_info->trans_block_rsv;
2530 * If we are using the NO_HOLES feature we might have had already an
2531 * hole that overlaps a part of the region [lockstart, lockend] and
2532 * ends at (or beyond) lockend. Since we have no file extent items to
2533 * represent holes, drop_end can be less than lockend and so we must
2534 * make sure we have an extent map representing the existing hole (the
2535 * call to __btrfs_drop_extents() might have dropped the existing extent
2536 * map representing the existing hole), otherwise the fast fsync path
2537 * will not record the existence of the hole region
2538 * [existing_hole_start, lockend].
2540 if (drop_args.drop_end <= end)
2541 drop_args.drop_end = end + 1;
2543 * Don't insert file hole extent item if it's for a range beyond eof
2544 * (because it's useless) or if it represents a 0 bytes range (when
2545 * cur_offset == drop_end).
2547 if (!extent_info && cur_offset < ino_size &&
2548 cur_offset < drop_args.drop_end) {
2549 ret = fill_holes(trans, inode, path, cur_offset,
2550 drop_args.drop_end);
2552 /* Same comment as above. */
2553 btrfs_abort_transaction(trans, ret);
2556 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2557 /* See the comment in the loop above for the reasoning here. */
2558 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2559 drop_args.drop_end - cur_offset);
2561 btrfs_abort_transaction(trans, ret);
2567 ret = btrfs_insert_replace_extent(trans, inode, path,
2568 extent_info, extent_info->data_len,
2569 drop_args.bytes_found);
2571 btrfs_abort_transaction(trans, ret);
2580 trans->block_rsv = &fs_info->trans_block_rsv;
2582 btrfs_end_transaction(trans);
2586 btrfs_free_block_rsv(fs_info, rsv);
2591 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2593 struct inode *inode = file_inode(file);
2594 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2595 struct btrfs_root *root = BTRFS_I(inode)->root;
2596 struct extent_state *cached_state = NULL;
2597 struct btrfs_path *path;
2598 struct btrfs_trans_handle *trans = NULL;
2603 u64 orig_start = offset;
2607 bool truncated_block = false;
2608 bool updated_inode = false;
2610 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2612 ret = btrfs_wait_ordered_range(inode, offset, len);
2614 goto out_only_mutex;
2616 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2617 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2619 goto out_only_mutex;
2621 /* Already in a large hole */
2623 goto out_only_mutex;
2626 ret = file_modified(file);
2628 goto out_only_mutex;
2630 lockstart = round_up(offset, fs_info->sectorsize);
2631 lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2632 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2633 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2635 * We needn't truncate any block which is beyond the end of the file
2636 * because we are sure there is no data there.
2639 * Only do this if we are in the same block and we aren't doing the
2642 if (same_block && len < fs_info->sectorsize) {
2643 if (offset < ino_size) {
2644 truncated_block = true;
2645 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2650 goto out_only_mutex;
2653 /* zero back part of the first block */
2654 if (offset < ino_size) {
2655 truncated_block = true;
2656 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2658 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2663 /* Check the aligned pages after the first unaligned page,
2664 * if offset != orig_start, which means the first unaligned page
2665 * including several following pages are already in holes,
2666 * the extra check can be skipped */
2667 if (offset == orig_start) {
2668 /* after truncate page, check hole again */
2669 len = offset + len - lockstart;
2671 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2673 goto out_only_mutex;
2676 goto out_only_mutex;
2681 /* Check the tail unaligned part is in a hole */
2682 tail_start = lockend + 1;
2683 tail_len = offset + len - tail_start;
2685 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2686 if (unlikely(ret < 0))
2687 goto out_only_mutex;
2689 /* zero the front end of the last page */
2690 if (tail_start + tail_len < ino_size) {
2691 truncated_block = true;
2692 ret = btrfs_truncate_block(BTRFS_I(inode),
2693 tail_start + tail_len,
2696 goto out_only_mutex;
2701 if (lockend < lockstart) {
2703 goto out_only_mutex;
2706 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
2708 path = btrfs_alloc_path();
2714 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2715 lockend, NULL, &trans);
2716 btrfs_free_path(path);
2720 ASSERT(trans != NULL);
2721 inode_inc_iversion(inode);
2722 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2723 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2724 updated_inode = true;
2725 btrfs_end_transaction(trans);
2726 btrfs_btree_balance_dirty(fs_info);
2728 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2731 if (!updated_inode && truncated_block && !ret) {
2733 * If we only end up zeroing part of a page, we still need to
2734 * update the inode item, so that all the time fields are
2735 * updated as well as the necessary btrfs inode in memory fields
2736 * for detecting, at fsync time, if the inode isn't yet in the
2737 * log tree or it's there but not up to date.
2739 struct timespec64 now = inode_set_ctime_current(inode);
2741 inode_inc_iversion(inode);
2742 inode_set_mtime_to_ts(inode, now);
2743 trans = btrfs_start_transaction(root, 1);
2744 if (IS_ERR(trans)) {
2745 ret = PTR_ERR(trans);
2749 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2750 ret2 = btrfs_end_transaction(trans);
2755 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2759 /* Helper structure to record which range is already reserved */
2760 struct falloc_range {
2761 struct list_head list;
2767 * Helper function to add falloc range
2769 * Caller should have locked the larger range of extent containing
2772 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2774 struct falloc_range *range = NULL;
2776 if (!list_empty(head)) {
2778 * As fallocate iterates by bytenr order, we only need to check
2781 range = list_last_entry(head, struct falloc_range, list);
2782 if (range->start + range->len == start) {
2788 range = kmalloc(sizeof(*range), GFP_KERNEL);
2791 range->start = start;
2793 list_add_tail(&range->list, head);
2797 static int btrfs_fallocate_update_isize(struct inode *inode,
2801 struct btrfs_trans_handle *trans;
2802 struct btrfs_root *root = BTRFS_I(inode)->root;
2806 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2809 trans = btrfs_start_transaction(root, 1);
2811 return PTR_ERR(trans);
2813 inode_set_ctime_current(inode);
2814 i_size_write(inode, end);
2815 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2816 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2817 ret2 = btrfs_end_transaction(trans);
2819 return ret ? ret : ret2;
2823 RANGE_BOUNDARY_WRITTEN_EXTENT,
2824 RANGE_BOUNDARY_PREALLOC_EXTENT,
2825 RANGE_BOUNDARY_HOLE,
2828 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2831 const u64 sectorsize = inode->root->fs_info->sectorsize;
2832 struct extent_map *em;
2835 offset = round_down(offset, sectorsize);
2836 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
2840 if (em->block_start == EXTENT_MAP_HOLE)
2841 ret = RANGE_BOUNDARY_HOLE;
2842 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2843 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2845 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2847 free_extent_map(em);
2851 static int btrfs_zero_range(struct inode *inode,
2856 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2857 struct extent_map *em;
2858 struct extent_changeset *data_reserved = NULL;
2861 const u64 sectorsize = fs_info->sectorsize;
2862 u64 alloc_start = round_down(offset, sectorsize);
2863 u64 alloc_end = round_up(offset + len, sectorsize);
2864 u64 bytes_to_reserve = 0;
2865 bool space_reserved = false;
2867 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2868 alloc_end - alloc_start);
2875 * Avoid hole punching and extent allocation for some cases. More cases
2876 * could be considered, but these are unlikely common and we keep things
2877 * as simple as possible for now. Also, intentionally, if the target
2878 * range contains one or more prealloc extents together with regular
2879 * extents and holes, we drop all the existing extents and allocate a
2880 * new prealloc extent, so that we get a larger contiguous disk extent.
2882 if (em->start <= alloc_start &&
2883 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2884 const u64 em_end = em->start + em->len;
2886 if (em_end >= offset + len) {
2888 * The whole range is already a prealloc extent,
2889 * do nothing except updating the inode's i_size if
2892 free_extent_map(em);
2893 ret = btrfs_fallocate_update_isize(inode, offset + len,
2898 * Part of the range is already a prealloc extent, so operate
2899 * only on the remaining part of the range.
2901 alloc_start = em_end;
2902 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2903 len = offset + len - alloc_start;
2904 offset = alloc_start;
2905 alloc_hint = em->block_start + em->len;
2907 free_extent_map(em);
2909 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2910 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2911 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2918 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2919 free_extent_map(em);
2920 ret = btrfs_fallocate_update_isize(inode, offset + len,
2924 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2925 free_extent_map(em);
2926 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2929 ret = btrfs_fallocate_update_isize(inode,
2934 free_extent_map(em);
2935 alloc_start = round_down(offset, sectorsize);
2936 alloc_end = alloc_start + sectorsize;
2940 alloc_start = round_up(offset, sectorsize);
2941 alloc_end = round_down(offset + len, sectorsize);
2944 * For unaligned ranges, check the pages at the boundaries, they might
2945 * map to an extent, in which case we need to partially zero them, or
2946 * they might map to a hole, in which case we need our allocation range
2949 if (!IS_ALIGNED(offset, sectorsize)) {
2950 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2954 if (ret == RANGE_BOUNDARY_HOLE) {
2955 alloc_start = round_down(offset, sectorsize);
2957 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2958 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2966 if (!IS_ALIGNED(offset + len, sectorsize)) {
2967 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2971 if (ret == RANGE_BOUNDARY_HOLE) {
2972 alloc_end = round_up(offset + len, sectorsize);
2974 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2975 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2985 if (alloc_start < alloc_end) {
2986 struct extent_state *cached_state = NULL;
2987 const u64 lockstart = alloc_start;
2988 const u64 lockend = alloc_end - 1;
2990 bytes_to_reserve = alloc_end - alloc_start;
2991 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2995 space_reserved = true;
2996 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2998 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
2999 alloc_start, bytes_to_reserve);
3001 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
3002 lockend, &cached_state);
3005 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3006 alloc_end - alloc_start,
3008 offset + len, &alloc_hint);
3009 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3011 /* btrfs_prealloc_file_range releases reserved space on error */
3013 space_reserved = false;
3017 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3019 if (ret && space_reserved)
3020 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3021 alloc_start, bytes_to_reserve);
3022 extent_changeset_free(data_reserved);
3027 static long btrfs_fallocate(struct file *file, int mode,
3028 loff_t offset, loff_t len)
3030 struct inode *inode = file_inode(file);
3031 struct extent_state *cached_state = NULL;
3032 struct extent_changeset *data_reserved = NULL;
3033 struct falloc_range *range;
3034 struct falloc_range *tmp;
3035 LIST_HEAD(reserve_list);
3043 u64 data_space_needed = 0;
3044 u64 data_space_reserved = 0;
3045 u64 qgroup_reserved = 0;
3046 struct extent_map *em;
3047 int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3050 /* Do not allow fallocate in ZONED mode */
3051 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3054 alloc_start = round_down(offset, blocksize);
3055 alloc_end = round_up(offset + len, blocksize);
3056 cur_offset = alloc_start;
3058 /* Make sure we aren't being give some crap mode */
3059 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3060 FALLOC_FL_ZERO_RANGE))
3063 if (mode & FALLOC_FL_PUNCH_HOLE)
3064 return btrfs_punch_hole(file, offset, len);
3066 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3068 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3069 ret = inode_newsize_ok(inode, offset + len);
3074 ret = file_modified(file);
3079 * TODO: Move these two operations after we have checked
3080 * accurate reserved space, or fallocate can still fail but
3081 * with page truncated or size expanded.
3083 * But that's a minor problem and won't do much harm BTW.
3085 if (alloc_start > inode->i_size) {
3086 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3090 } else if (offset + len > inode->i_size) {
3092 * If we are fallocating from the end of the file onward we
3093 * need to zero out the end of the block if i_size lands in the
3094 * middle of a block.
3096 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3102 * We have locked the inode at the VFS level (in exclusive mode) and we
3103 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3104 * locking the file range, flush all dealloc in the range and wait for
3105 * all ordered extents in the range to complete. After this we can lock
3106 * the file range and, due to the previous locking we did, we know there
3107 * can't be more delalloc or ordered extents in the range.
3109 ret = btrfs_wait_ordered_range(inode, alloc_start,
3110 alloc_end - alloc_start);
3114 if (mode & FALLOC_FL_ZERO_RANGE) {
3115 ret = btrfs_zero_range(inode, offset, len, mode);
3116 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3120 locked_end = alloc_end - 1;
3121 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3124 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3126 /* First, check if we exceed the qgroup limit */
3127 while (cur_offset < alloc_end) {
3128 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3129 alloc_end - cur_offset);
3134 last_byte = min(extent_map_end(em), alloc_end);
3135 actual_end = min_t(u64, extent_map_end(em), offset + len);
3136 last_byte = ALIGN(last_byte, blocksize);
3137 if (em->block_start == EXTENT_MAP_HOLE ||
3138 (cur_offset >= inode->i_size &&
3139 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3140 const u64 range_len = last_byte - cur_offset;
3142 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3144 free_extent_map(em);
3147 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3148 &data_reserved, cur_offset, range_len);
3150 free_extent_map(em);
3153 qgroup_reserved += range_len;
3154 data_space_needed += range_len;
3156 free_extent_map(em);
3157 cur_offset = last_byte;
3160 if (!ret && data_space_needed > 0) {
3162 * We are safe to reserve space here as we can't have delalloc
3163 * in the range, see above.
3165 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3168 data_space_reserved = data_space_needed;
3172 * If ret is still 0, means we're OK to fallocate.
3173 * Or just cleanup the list and exit.
3175 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3177 ret = btrfs_prealloc_file_range(inode, mode,
3179 range->len, i_blocksize(inode),
3180 offset + len, &alloc_hint);
3182 * btrfs_prealloc_file_range() releases space even
3183 * if it returns an error.
3185 data_space_reserved -= range->len;
3186 qgroup_reserved -= range->len;
3187 } else if (data_space_reserved > 0) {
3188 btrfs_free_reserved_data_space(BTRFS_I(inode),
3189 data_reserved, range->start,
3191 data_space_reserved -= range->len;
3192 qgroup_reserved -= range->len;
3193 } else if (qgroup_reserved > 0) {
3194 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3195 range->start, range->len, NULL);
3196 qgroup_reserved -= range->len;
3198 list_del(&range->list);
3205 * We didn't need to allocate any more space, but we still extended the
3206 * size of the file so we need to update i_size and the inode item.
3208 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3210 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3213 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3214 extent_changeset_free(data_reserved);
3219 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3220 * that has unflushed and/or flushing delalloc. There might be other adjacent
3221 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3222 * looping while it gets adjacent subranges, and merging them together.
3224 static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3225 struct extent_state **cached_state,
3226 bool *search_io_tree,
3227 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3229 u64 len = end + 1 - start;
3230 u64 delalloc_len = 0;
3231 struct btrfs_ordered_extent *oe;
3236 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3237 * means we have delalloc (dirty pages) for which writeback has not
3240 if (*search_io_tree) {
3241 spin_lock(&inode->lock);
3242 if (inode->delalloc_bytes > 0) {
3243 spin_unlock(&inode->lock);
3244 *delalloc_start_ret = start;
3245 delalloc_len = count_range_bits(&inode->io_tree,
3246 delalloc_start_ret, end,
3247 len, EXTENT_DELALLOC, 1,
3250 spin_unlock(&inode->lock);
3254 if (delalloc_len > 0) {
3256 * If delalloc was found then *delalloc_start_ret has a sector size
3257 * aligned value (rounded down).
3259 *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3261 if (*delalloc_start_ret == start) {
3262 /* Delalloc for the whole range, nothing more to do. */
3263 if (*delalloc_end_ret == end)
3265 /* Else trim our search range for ordered extents. */
3266 start = *delalloc_end_ret + 1;
3267 len = end + 1 - start;
3270 /* No delalloc, future calls don't need to search again. */
3271 *search_io_tree = false;
3275 * Now also check if there's any ordered extent in the range.
3276 * We do this because:
3278 * 1) When delalloc is flushed, the file range is locked, we clear the
3279 * EXTENT_DELALLOC bit from the io tree and create an extent map and
3280 * an ordered extent for the write. So we might just have been called
3281 * after delalloc is flushed and before the ordered extent completes
3282 * and inserts the new file extent item in the subvolume's btree;
3284 * 2) We may have an ordered extent created by flushing delalloc for a
3285 * subrange that starts before the subrange we found marked with
3286 * EXTENT_DELALLOC in the io tree.
3288 * We could also use the extent map tree to find such delalloc that is
3289 * being flushed, but using the ordered extents tree is more efficient
3290 * because it's usually much smaller as ordered extents are removed from
3291 * the tree once they complete. With the extent maps, we mau have them
3292 * in the extent map tree for a very long time, and they were either
3293 * created by previous writes or loaded by read operations.
3295 oe = btrfs_lookup_first_ordered_range(inode, start, len);
3297 return (delalloc_len > 0);
3299 /* The ordered extent may span beyond our search range. */
3300 oe_start = max(oe->file_offset, start);
3301 oe_end = min(oe->file_offset + oe->num_bytes - 1, end);
3303 btrfs_put_ordered_extent(oe);
3305 /* Don't have unflushed delalloc, return the ordered extent range. */
3306 if (delalloc_len == 0) {
3307 *delalloc_start_ret = oe_start;
3308 *delalloc_end_ret = oe_end;
3313 * We have both unflushed delalloc (io_tree) and an ordered extent.
3314 * If the ranges are adjacent returned a combined range, otherwise
3315 * return the leftmost range.
3317 if (oe_start < *delalloc_start_ret) {
3318 if (oe_end < *delalloc_start_ret)
3319 *delalloc_end_ret = oe_end;
3320 *delalloc_start_ret = oe_start;
3321 } else if (*delalloc_end_ret + 1 == oe_start) {
3322 *delalloc_end_ret = oe_end;
3329 * Check if there's delalloc in a given range.
3331 * @inode: The inode.
3332 * @start: The start offset of the range. It does not need to be
3333 * sector size aligned.
3334 * @end: The end offset (inclusive value) of the search range.
3335 * It does not need to be sector size aligned.
3336 * @cached_state: Extent state record used for speeding up delalloc
3337 * searches in the inode's io_tree. Can be NULL.
3338 * @delalloc_start_ret: Output argument, set to the start offset of the
3339 * subrange found with delalloc (may not be sector size
3341 * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3342 * of the subrange found with delalloc.
3344 * Returns true if a subrange with delalloc is found within the given range, and
3345 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3346 * end offsets of the subrange.
3348 bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3349 struct extent_state **cached_state,
3350 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3352 u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3353 u64 prev_delalloc_end = 0;
3354 bool search_io_tree = true;
3357 while (cur_offset <= end) {
3362 delalloc = find_delalloc_subrange(inode, cur_offset, end,
3363 cached_state, &search_io_tree,
3369 if (prev_delalloc_end == 0) {
3370 /* First subrange found. */
3371 *delalloc_start_ret = max(delalloc_start, start);
3372 *delalloc_end_ret = delalloc_end;
3374 } else if (delalloc_start == prev_delalloc_end + 1) {
3375 /* Subrange adjacent to the previous one, merge them. */
3376 *delalloc_end_ret = delalloc_end;
3378 /* Subrange not adjacent to the previous one, exit. */
3382 prev_delalloc_end = delalloc_end;
3383 cur_offset = delalloc_end + 1;
3391 * Check if there's a hole or delalloc range in a range representing a hole (or
3392 * prealloc extent) found in the inode's subvolume btree.
3394 * @inode: The inode.
3395 * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3396 * @start: Start offset of the hole region. It does not need to be sector
3398 * @end: End offset (inclusive value) of the hole region. It does not
3399 * need to be sector size aligned.
3400 * @start_ret: Return parameter, used to set the start of the subrange in the
3401 * hole that matches the search criteria (seek mode), if such
3402 * subrange is found (return value of the function is true).
3403 * The value returned here may not be sector size aligned.
3405 * Returns true if a subrange matching the given seek mode is found, and if one
3406 * is found, it updates @start_ret with the start of the subrange.
3408 static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3409 struct extent_state **cached_state,
3410 u64 start, u64 end, u64 *start_ret)
3416 delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state,
3417 &delalloc_start, &delalloc_end);
3418 if (delalloc && whence == SEEK_DATA) {
3419 *start_ret = delalloc_start;
3423 if (delalloc && whence == SEEK_HOLE) {
3425 * We found delalloc but it starts after out start offset. So we
3426 * have a hole between our start offset and the delalloc start.
3428 if (start < delalloc_start) {
3433 * Delalloc range starts at our start offset.
3434 * If the delalloc range's length is smaller than our range,
3435 * then it means we have a hole that starts where the delalloc
3438 if (delalloc_end < end) {
3439 *start_ret = delalloc_end + 1;
3443 /* There's delalloc for the whole range. */
3447 if (!delalloc && whence == SEEK_HOLE) {
3453 * No delalloc in the range and we are seeking for data. The caller has
3454 * to iterate to the next extent item in the subvolume btree.
3459 static loff_t find_desired_extent(struct file *file, loff_t offset, int whence)
3461 struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host);
3462 struct btrfs_file_private *private = file->private_data;
3463 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3464 struct extent_state *cached_state = NULL;
3465 struct extent_state **delalloc_cached_state;
3466 const loff_t i_size = i_size_read(&inode->vfs_inode);
3467 const u64 ino = btrfs_ino(inode);
3468 struct btrfs_root *root = inode->root;
3469 struct btrfs_path *path;
3470 struct btrfs_key key;
3471 u64 last_extent_end;
3478 if (i_size == 0 || offset >= i_size)
3482 * Quick path. If the inode has no prealloc extents and its number of
3483 * bytes used matches its i_size, then it can not have holes.
3485 if (whence == SEEK_HOLE &&
3486 !(inode->flags & BTRFS_INODE_PREALLOC) &&
3487 inode_get_bytes(&inode->vfs_inode) == i_size)
3491 private = kzalloc(sizeof(*private), GFP_KERNEL);
3493 * No worries if memory allocation failed.
3494 * The private structure is used only for speeding up multiple
3495 * lseek SEEK_HOLE/DATA calls to a file when there's delalloc,
3496 * so everything will still be correct.
3498 file->private_data = private;
3502 delalloc_cached_state = &private->llseek_cached_state;
3504 delalloc_cached_state = NULL;
3507 * offset can be negative, in this case we start finding DATA/HOLE from
3508 * the very start of the file.
3510 start = max_t(loff_t, 0, offset);
3512 lockstart = round_down(start, fs_info->sectorsize);
3513 lockend = round_up(i_size, fs_info->sectorsize);
3514 if (lockend <= lockstart)
3515 lockend = lockstart + fs_info->sectorsize;
3518 path = btrfs_alloc_path();
3521 path->reada = READA_FORWARD;
3524 key.type = BTRFS_EXTENT_DATA_KEY;
3527 last_extent_end = lockstart;
3529 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3531 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3534 } else if (ret > 0 && path->slots[0] > 0) {
3535 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3536 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3540 while (start < i_size) {
3541 struct extent_buffer *leaf = path->nodes[0];
3542 struct btrfs_file_extent_item *extent;
3546 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3547 ret = btrfs_next_leaf(root, path);
3553 leaf = path->nodes[0];
3556 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3557 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3560 extent_end = btrfs_file_extent_end(path);
3563 * In the first iteration we may have a slot that points to an
3564 * extent that ends before our start offset, so skip it.
3566 if (extent_end <= start) {
3571 /* We have an implicit hole, NO_HOLES feature is likely set. */
3572 if (last_extent_end < key.offset) {
3573 u64 search_start = last_extent_end;
3577 * First iteration, @start matches @offset and it's
3580 if (start == offset)
3581 search_start = offset;
3583 found = find_desired_extent_in_hole(inode, whence,
3584 delalloc_cached_state,
3589 start = found_start;
3593 * Didn't find data or a hole (due to delalloc) in the
3594 * implicit hole range, so need to analyze the extent.
3598 extent = btrfs_item_ptr(leaf, path->slots[0],
3599 struct btrfs_file_extent_item);
3600 type = btrfs_file_extent_type(leaf, extent);
3603 * Can't access the extent's disk_bytenr field if this is an
3604 * inline extent, since at that offset, it's where the extent
3607 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3608 (type == BTRFS_FILE_EXTENT_REG &&
3609 btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3611 * Explicit hole or prealloc extent, search for delalloc.
3612 * A prealloc extent is treated like a hole.
3614 u64 search_start = key.offset;
3618 * First iteration, @start matches @offset and it's
3621 if (start == offset)
3622 search_start = offset;
3624 found = find_desired_extent_in_hole(inode, whence,
3625 delalloc_cached_state,
3630 start = found_start;
3634 * Didn't find data or a hole (due to delalloc) in the
3635 * implicit hole range, so need to analyze the next
3640 * Found a regular or inline extent.
3641 * If we are seeking for data, adjust the start offset
3642 * and stop, we're done.
3644 if (whence == SEEK_DATA) {
3645 start = max_t(u64, key.offset, offset);
3650 * Else, we are seeking for a hole, check the next file
3656 last_extent_end = extent_end;
3658 if (fatal_signal_pending(current)) {
3665 /* We have an implicit hole from the last extent found up to i_size. */
3666 if (!found && start < i_size) {
3667 found = find_desired_extent_in_hole(inode, whence,
3668 delalloc_cached_state, start,
3669 i_size - 1, &start);
3675 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3676 btrfs_free_path(path);
3681 if (whence == SEEK_DATA && start >= i_size)
3684 return min_t(loff_t, start, i_size);
3687 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3689 struct inode *inode = file->f_mapping->host;
3693 return generic_file_llseek(file, offset, whence);
3696 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3697 offset = find_desired_extent(file, offset, whence);
3698 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3705 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3708 static int btrfs_file_open(struct inode *inode, struct file *filp)
3712 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC |
3715 ret = fsverity_file_open(inode, filp);
3718 return generic_file_open(inode, filp);
3721 static int check_direct_read(struct btrfs_fs_info *fs_info,
3722 const struct iov_iter *iter, loff_t offset)
3727 ret = check_direct_IO(fs_info, iter, offset);
3731 if (!iter_is_iovec(iter))
3734 for (seg = 0; seg < iter->nr_segs; seg++) {
3735 for (i = seg + 1; i < iter->nr_segs; i++) {
3736 const struct iovec *iov1 = iter_iov(iter) + seg;
3737 const struct iovec *iov2 = iter_iov(iter) + i;
3739 if (iov1->iov_base == iov2->iov_base)
3746 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3748 struct inode *inode = file_inode(iocb->ki_filp);
3749 size_t prev_left = 0;
3753 if (fsverity_active(inode))
3756 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3759 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3762 * This is similar to what we do for direct IO writes, see the comment
3763 * at btrfs_direct_write(), but we also disable page faults in addition
3764 * to disabling them only at the iov_iter level. This is because when
3765 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3766 * which can still trigger page fault ins despite having set ->nofault
3767 * to true of our 'to' iov_iter.
3769 * The difference to direct IO writes is that we deadlock when trying
3770 * to lock the extent range in the inode's tree during he page reads
3771 * triggered by the fault in (while for writes it is due to waiting for
3772 * our own ordered extent). This is because for direct IO reads,
3773 * btrfs_dio_iomap_begin() returns with the extent range locked, which
3774 * is only unlocked in the endio callback (end_bio_extent_readpage()).
3776 pagefault_disable();
3778 ret = btrfs_dio_read(iocb, to, read);
3779 to->nofault = false;
3782 /* No increment (+=) because iomap returns a cumulative value. */
3786 if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3787 const size_t left = iov_iter_count(to);
3789 if (left == prev_left) {
3791 * We didn't make any progress since the last attempt,
3792 * fallback to a buffered read for the remainder of the
3793 * range. This is just to avoid any possibility of looping
3799 * We made some progress since the last retry or this is
3800 * the first time we are retrying. Fault in as many pages
3801 * as possible and retry.
3803 fault_in_iov_iter_writeable(to, left);
3808 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3809 return ret < 0 ? ret : read;
3812 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3816 if (iocb->ki_flags & IOCB_DIRECT) {
3817 ret = btrfs_direct_read(iocb, to);
3818 if (ret < 0 || !iov_iter_count(to) ||
3819 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3823 return filemap_read(iocb, to, ret);
3826 const struct file_operations btrfs_file_operations = {
3827 .llseek = btrfs_file_llseek,
3828 .read_iter = btrfs_file_read_iter,
3829 .splice_read = filemap_splice_read,
3830 .write_iter = btrfs_file_write_iter,
3831 .splice_write = iter_file_splice_write,
3832 .mmap = btrfs_file_mmap,
3833 .open = btrfs_file_open,
3834 .release = btrfs_release_file,
3835 .get_unmapped_area = thp_get_unmapped_area,
3836 .fsync = btrfs_sync_file,
3837 .fallocate = btrfs_fallocate,
3838 .unlocked_ioctl = btrfs_ioctl,
3839 #ifdef CONFIG_COMPAT
3840 .compat_ioctl = btrfs_compat_ioctl,
3842 .remap_file_range = btrfs_remap_file_range,
3845 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3850 * So with compression we will find and lock a dirty page and clear the
3851 * first one as dirty, setup an async extent, and immediately return
3852 * with the entire range locked but with nobody actually marked with
3853 * writeback. So we can't just filemap_write_and_wait_range() and
3854 * expect it to work since it will just kick off a thread to do the
3855 * actual work. So we need to call filemap_fdatawrite_range _again_
3856 * since it will wait on the page lock, which won't be unlocked until
3857 * after the pages have been marked as writeback and so we're good to go
3858 * from there. We have to do this otherwise we'll miss the ordered
3859 * extents and that results in badness. Please Josef, do not think you
3860 * know better and pull this out at some point in the future, it is
3861 * right and you are wrong.
3863 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3864 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3865 &BTRFS_I(inode)->runtime_flags))
3866 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);